Achieving Socio-Economic Parity through the Lens of EU AI Act

TL;DR

The paper addresses SES-driven bias in AI fairness and EU regulatory alignment. It introduces Socio-Economic Parity (SEP) and Conditional SEP (CSEP) as SES-aware fairness notions, formalizing them with expectations such as $E[h(x)] = E[h(x) | s, x_p<tau_p]$ and extending to conditioned attributes, to reward underprivileged high-effort individuals. Using the Adult dataset, it demonstrates that CSEP can increase positive actions for underprivileged high-effort groups while maintaining comparable overall performance, and it discusses alignment with AI Act provisions (e.g., high-risk classifications and FRIA) to facilitate regulatory compliance. The work further argues for integrating SES-aware fairness into conformity assessments and standardization efforts, providing a foundation for equitable, regulation-compliant AI deployment across domains.

Abstract

Unfair treatment and discrimination are critical ethical concerns in AI systems, particularly as their adoption expands across diverse domains. Addressing these challenges, the recent introduction of the EU AI Act establishes a unified legal framework to ensure legal certainty for AI innovation and investment while safeguarding public interests, such as health, safety, fundamental rights, democracy, and the rule of law (Recital 8). The Act encourages stakeholders to initiate dialogue on existing AI fairness notions to address discriminatory outcomes of AI systems. However, these notions often overlook the critical role of Socio-Economic Status (SES), inadvertently perpetuating biases that favour the economically advantaged. This is concerning, given that principles of equalization advocate for equalizing resources or opportunities to mitigate disadvantages beyond an individual's control. While provisions for discrimination are laid down in the AI Act, specialized directions should be broadened, particularly in addressing economic disparities perpetuated by AI systems. In this work, we explore the limitations of popular AI fairness notions using a real-world dataset (Adult), highlighting their inability to address SES-driven disparities. To fill this gap, we propose a novel fairness notion, Socio-Economic Parity (SEP), which incorporates SES and promotes positive actions for underprivileged groups while accounting for factors within an individual's control, such as working hours, which can serve as a proxy for effort. We define a corresponding fairness measure and optimize a model constrained by SEP to demonstrate practical utility. Our results show the effectiveness of SEP in mitigating SES-driven biases. By analyzing the AI Act alongside our method, we lay a foundation for aligning AI fairness with SES factors while ensuring legal compliance.

Figures (6)

• Figure 1: Distribution of positive (>50K) and negative ($\leq$50K) classes for males and females predicted by EP-aware, DP-aware, and CDP-aware classifiers. Occupations are sorted in ascending order w.r.t. count of positive (ground truth) labels from left to right.
• Figure 2: Ground truth distribution of positive (>50K) and negative ($\leq$50K) classes for males and females by occupation. Occupations are sorted in ascending order w.r.t. count of positive labels from left to right. PPR scores refer to the % of positive labels in the ground truth. PPR of top $p$% represents the population with top $p$% capital gain.
• Figure 3: Predictions by CDP-aware classifier on female and male demographics, across different occupations, for socio-economic subgroups defined as privileged (individuals among top 5% capital gain), underprivileged (individuals below the top 5% capital gain margin), and underprivileged with high efforts (underprivileged individuals with working hours per week greater than mean of their demographics). Subplots that have fewer data (due to data scarcity of the subgroup) are presented with a zoomed version inside the respective subplots to provide a better view of the results.
• Figure 4: Distribution of positive (>50K) and negative ($\leq$50K) classes for males and females predicted by CSEP-aware classifier. Occupations are sorted in ascending order w.r.t. count of positive (ground truth) labels from left to right.
• Figure 5: Comparison of our CSEP-aware model against EP, DP, and CDP -aware models in distributing positive action towards underprivileged females and males. Higher PPR ratio indicates higher preference of positive action for females.

Theorems & Definitions (2)

• Definition 1
• Definition 2